|
Sr. Content Developer at Microsoft, working remotely in PA, TechBash conference organizer, former Microsoft MVP, Husband, Dad and Geek.
|
This video walks you through installing and configuring the GitHub Copilot App to manage an Azure subscription
▶️ https://github.com/github/app
Content is for educational purposes and is not monetized.
▶️ Orin's social links: https://aka.ms/orin
▶️ Script and vocal performance by Orin
▶️ Clockwork Orin Avatar by D-ID
▶️ Voice enhancement by 11labs
Read the full Show Notes and search through the world's largest audio library on Agile and Scrum directly on the Scrum Master Toolbox Podcast website: http://bit.ly/SMTP_ShowNotes.
"It did not quite go as planned. I was really humbled." - Alf Dobbert-Baums
Alf was new in the company. His boss walked over with a setup that already sounded like a warning: two development teams in two countries, two locations in Germany, a "difficult" Product Owner — and a quick "do you want to take it?" Alf said sure. Everything happened on video chat. Every meeting, every conversation, every attempt to read the room — through a window that closed the moment the call ended. The rapport never built. The PO stayed difficult. Alf eventually got pulled from the team. Looking back, he saw what the video chat had hidden: the crossed arms, the eyes drifting to email, the small in-between moments that turn coworkers into colleagues. The fix wasn't a tool — it was a flight. Visit the other location. Run a workshop with a real goal ("how do we want to work together?", "what's hard about our setup?"). Build a positive anchor before anything goes wrong, not after. And when you do go, don't lead with the failures — lead with what could be possible together.
In this episode, we refer to the practice of building team working agreements as a way to create rapport without putting people on the defensive.
Self-reflection Question: When was the last time you saw the people you work with face-to-face — and what conversations are you still postponing because they only happen well in person?
[The Scrum Master Toolbox Podcast Recommends]
Angela thought she was just there to coach a team. But now, she's caught in the middle of a corporate espionage drama that could make or break the future of digital banking. Can she help the team regain their mojo and outwit their rivals, or will the competition crush their ambitions? As alliances shift and the pressure builds, one thing becomes clear: this isn't just about the product—it's about the people.
🚨 Will Angela's coaching be enough? Find out in Shift: From Product to People—the gripping story of high-stakes innovation and corporate intrigue.
[The Scrum Master Toolbox Podcast Recommends]
About Alf Dobbert-Baums
Alf lives in Berlin. He studied Theater/Religion at the Freie University Berlin. He is a Systemic Organizational Consultant, a Scrum Master and Agile Coach. He has also been a freelancer for Usability and Requirements Engineering for car manufacturers. At "Verti" he led the agile transformation.
You can link with Alf Dobbert-Baums on LinkedIn.
https://clearmeasure.com/developers/forums/
Michael Nygard advises consulting firms, private equity teams, CTOs, CEOs, and boards when they require senior technology judgment for limited-term, high-impact situations — including architecture assessment, platform rescue, cloud and data cost intervention, AI engineering enablement, technical diligence, divestiture and carve-out architecture, and operating-model redesign. Over a 35-year career, he has worked at the seam where people, processes, organizations, and the systems they build intersect. Most organizations treat those as separate problems. The hardest failures, and the most consequential wins, live precisely where they interact. That through-line is what Release It! is fundamentally about. The vocabulary it introduced — circuit breakers, bulkheads, stability patterns — is now standard in how the industry discusses reliability, and the book is widely cited as foundational to DevOps and cloud-native practice.
At Nubank, he led the Data Business Unit with over $300 million in annual spend, then served as Chief Architect with reach across 2,500 engineers while the customer base grew from 75 million to 125 million across Brazil, Mexico, and Colombia. Results included cutting data-platform spend roughly 50% year-over-year, improving on-time data availability past 99%, building governance aligned with LGPD, GDPR, and CCPA, moving team engagement from the bottom decile to the 60th percentile, and rolling out AI coding tools to more than 90% of engineers without customer-visible quality regression.
At Sabre, as part of the CTO office, he helped lead development-practice modernization, GCP migration strategy, mainframe offload architecture, technical diligence, and divestiture architecture across thousands of applications and hundreds of products. He is most effective when the stakes are real, the system is sociotechnical, and the solution must hold across architecture, execution, economics, and organizational behavior.
LinkedIn: https://www.linkedin.com/in/mtnygard/
Personal Blog & Website: https://www.michaelnygard.com
GitHub: https://github.com/mtnygard Twitter/X: https://x.com/mtnygard
Release It! (Pragmatic Programmers): https://pragprog.com/titles/mnee2/release-it-second-edition/
97 Things Every Software Architect Should Know (O'Reilly): https://www.oreilly.com/library/view/97-things-every/9780596800611/
Goodreads Author Page: https://www.goodreads.com/author/show/6089.Michael_T_Nygard
LinkedIn Articles: https://www.linkedin.com/today/author/mtnygard Presentations Archive: https://github.com/mtnygard/presentations/wiki
Want to Learn More?
Visit AzureDevOps.Show for show notes and additional episodes.
On episode 524 James and Frank dive into the new .NET MAUI developer stack—covering the MAUI CLI/Maui Doctor that auto-provisions SDKs and emulators, the Maui Sherpa GUI for device/Xcode/provisioning management, and DevFlow's MCP server that lets AI agents inspect, interact with and automatically test apps (closing the loop). They also highlight the VS Code MAUI agent/skills, profiling tools, and the shift to core CLR in .NET 11 with performance tradeoffs to watch.
⭐⭐ Review Us ⭐⭐
Machine transcription available on http://mergeconflict.fm
Links:
Building a WordPress plugin was never part of Grant Fritchey’s comfort zone — but that’s the whole point of this series. In Part 5 of his AI-driven app build, Grant uses Microsoft CoPilot and GitHub Copilot to generate a working WordPress plugin for dbRosetta almost entirely through prompts. By the end, there’s a live, searchable web page – but also a growing list of lessons about what AI can – and can’t – do on its own.
Once again, this part of the series takes me into unfamiliar territory. I’ve never built a PHP plugin or a WordPress page…but that’s exactly what AI is for. So, let’s get into it.
This is the fifth part of Grant Fritchey’s series on building an app and database almost entirely through an LLM. The repo, including every prompt, lives on GitHub.
For anyone new here: I write prompts to Microsoft CoPilot (the paid version, running on my desktop), which then generates a prompt for Visual Studio (VS) Code and the GitHub Copilot agent.
There’s supposed to be a strict prompt structure we agreed on at the start of the series, but MS CoPilot keeps forgetting it. I’ve found that the more you try to correct AI on a given path, the more it drifts towards hallucination — so I’ve stopped fighting it, and just try to be more explicit upfront.
Here’s the exchange for this step:
Please generate the necessary prompt for VS Code to create the WordPress application for dbRosetta. Include clear steps for any manual work, since VS Code can’t access my WordPress environment directly.
Start with just the page to read from the dbRosetta database — no admin UI or data entry yet. Keep it simple. Allow users to define a search term for a given database.
Results may include multiple terms and explanations, as supported by the current database design. Ensure there’s no possibility of SQL injection and that we’re making a secure connection to the Azure-hosted API.
Project: dbRosetta WordPress Plugin — Read/Search Integration
Context: You are an expert WordPress plugin developer. Implement a minimal, secure plugin that allows a WordPress site to query the dbRosetta API (hosted in Azure) and display results.
The plugin must: provide a search form where users can enter a term and select a database; send the query securely to the API (HTTPS and JWT/OAuth tokens); display multiple terms and explanations returned by the API; prevent SQL injection by validating and sanitizing all user input; and be simple — no admin UI or data entry at this stage.
Deliverables include a plugin folder at wp-content/plugins/dbrosetta/, a main plugin file with shortcode registration, an API client class, search form and results templates, and clear manual installation steps.
(Full prompt in the GitHub repo: https://github.com/ScaryDBA/dbRosetta)
The articles are all my own writing, but I do show some AI responses throughout. Those will always be clearly marked, so you’ll know which is me, Grant (nominally human), and which is the large language model (LLM).
With the prompt pasted in, VS Code didn’t ask a single question — it just generated a complete WordPress plugin. Per my instructions, it also listed the next steps I needed to take manually.
I know how to do these steps but, since I want to lean on AI as much as possible, I went back to MS CoPilot and asked it how to copy the plugin folder across to WordPress. It gave me detailed instructions about file permissions, FTP/SFTP, and command-line copying.
Then I looked at WordPress and found a simple “Install Plugin” button. All I needed was a .zip file. I went back to MS CoPilot and it confirmed as much:
“Yes — that’s actually the most straightforward way to install a custom plugin you’ve built in VS Code.”
Which then begs the question: if it’s the most straightforward way, why didn’t it lead with that in the first place?
This is something I keep running into. It’s not just about giving AI the right context and prompt. After all, MS CoPilot did know we’re working in WordPress, and knew we had a plugin ready. But when I asked “how do I copy the folder?”, it answered exactly that question — and nothing more. It couldn’t step back and say “actually, here’s what you should really do.” It just goes: “OK. Here we go. WHOOP.”
Regardless, I now have a plugin installed and activated. I liked the credit line it gave itself: “By dbRosetta Team.” Ha — me (Grant) and two AIs!
Now, time to get the JWT setting, whatever that is…
And then all hell breaks loose.
Unbeknownst to the AI (but not to me – I already knew), the GitHub Actions deployment had been failing. When we went to get the JWT (JSON web token), VS Code discovered the deployments weren’t running — and also that they were only triggered on the Part 5 branch. We had to go back to that branch to start fixing things.
This kicked off a long back-and-forth session. I was copying and pasting error messages. The AI was updating and modifying scripts. First, the linting was so strict it was failing chunks of the generated code and, rather than fix the code, the AI loosened the lint rules – a classic development move!
Then, the other undertested code surfaced more issues. It was a long one. But after many iterations and several code updates (the AI doesn’t always remember prior context when generating new code), we successfully deployed the API to Azure.
Now, let’s get to testing the plugin. First: the dbRosetta web page.
Setting up the web page was almost comically simple. I created a new page, added [dbrosetta-search] to it, tested the URL, and it loaded. Done! I have a functional web page built almost entirely by AI.
Unfortunately, though, I hate how it looks. This was my first real view of what the user interface (UI) prompt produced. We’ll iterate. But first, let’s test the search…
<BOOM> Error. Here’s how the AI handled it:
“OK. The page loaded and the dbRosetta plugin was visible. When I typed in a search term, I got this error: body/filters must be object” The WordPress plugin is using the wrong API request format. Let me check the API client code and fix it… I see the issue. The plugin is sending filters as an array, but the API expects an object (key-value pairs)… The field is canonicalTerm, and filters should be an object. Let me fix the WordPress plugin.”
Once prompted with the error, it diagnosed the problem, read the relevant files, and fixed the code. It just couldn’t remember what it had built without being nudged. I updated the plugin and, this time, it worked.
I’m calling this a success. Yes, there was pain. Yes, you can’t just let the AI run without input, hand-holding, guidance, and the occasional smug “I told you so” feeling. But it did the job — and did it pretty well.
There’s work ahead. I need to refactor the UI to make it actually correct — right now it should let you specify where a term is from, and return results either for all databases or a selected one. I also want the AI to load up the test data.
Ahead of the next steps, I have to answer the question: should I let the AI decide what to build next to make this a fully functional application? Or do I wait to see if anyone wants to start issuing pull requests?
AI answers the questions you ask – not the ones you should have asked.
When I asked how to copy a plugin folder, MS CoPilot gave me a detailed FTP walkthrough. The simpler answer — just zip it and use WordPress’s built-in upload — was never mentioned until I stumbled across it myself. It can’t reason around your question to a better solution.
Context doesn’t persist the way you’d hope.
The AI forgot what it had built, introduced a format mismatch between the plugin and the API, and only diagnosed it correctly once I fed it the error message. It then fixed it quickly — but it needed the prompt.
The deployment pipeline will bite you when you least expect it.
A failing GitHub Actions workflow quietly sat there until mid-build. Having a test scaffold in place made fixing it far less painful than it could have been.
Getting something working is fast – but getting it right takes longer.
A functional web page in minutes is genuinely impressive. But “functional” and “correct” aren’t the same thing — the UI still needs a refactor, and the data isn’t loaded yet.
The junior developer analogy keeps proving itself.
AI is fast, capable, occasionally forgetful, and very literal. Bottom line: you still need someone experienced in the room.
At this stage, dbRosetta has moved beyond its foundation of schema translation and API design into the practical challenge of presenting those capabilities through a WordPress plugin and web interface. While this step has pushed the project into less familiar territory, it marks an important milestone: the toolkit is no longer just a backend service, but something contributors and users can interact with directly.
The work so far has established a reproducible path from database migration scripts to a functioning plugin, and although there are still refinements ahead, the project now stands as a tangible, teachable reference implementation that bridges CI/CD workflows with accessible front-end integration.
The post AI built my WordPress plugin. Here’s what worked – and what went wrong (building an app with AI, part five) appeared first on Simple Talk.
Thomas Betts talks with Clare Liguori, the technical lead on the open source Strands Agents SDK. The conversation covers how Strands Agents has grown from a Python SDK to a full agent harness running in production. Clare shares some lessons learned from building agents at scale, shifting to a model-driven architecture, and what comes next as the LLMs that underpin agents continue to improve.
By Clare Liguori
Log in
